AI chip startup d-Matrix is taking a major step toward the growing AI inference market through a new collaboration with Nvidia.
The startup announced that it will adopt Nvidia’s NVLink Fusion technology to integrate its Raptor processors directly into Nvidia’s data-centre systems.
The partnership highlights a broader shift taking place across the artificial intelligence industry: attention is moving beyond simply training AI models toward making them faster and cheaper to operate once they are deployed.
From Training AI to Running AI
AI infrastructure has traditionally focused heavily on training.
Training involves processing enormous datasets to create models capable of understanding language, images, code and other information.
But once those models have been trained, they still need to respond to users.
That process is known as inference.
Every time someone asks an AI assistant a question, requests an image, uses an AI coding tool or interacts with a voice agent, inference infrastructure is doing the work behind the scenes.
As AI adoption grows, the cost and speed of inference are becoming increasingly important.
This is where d-Matrix is positioning itself.
The company’s Raptor processors are designed specifically for inference workloads, where low latency and efficient data movement can make a major difference.
Nvidia Wants More Chips Inside Its Ecosystem
Nvidia has built one of the world’s most important AI computing platforms around its GPUs.
However, the company is also expanding the ecosystem around its hardware.
Its NVLink Fusion technology allows specialised processors from other companies to connect with Nvidia’s larger data-centre architecture.
For d-Matrix, this creates an opportunity to place its technology directly inside Nvidia-compatible systems rather than forcing customers to build completely separate infrastructure.
The company said its Raptor processors will connect into Nvidia server racks using NVLink Fusion, with the combined systems expected to become available in 2027.
Why This Matters for AI Companies
AI companies are increasingly looking for ways to reduce the cost of serving models.
Training a large model is expensive, but operating that model for millions of users can become an equally significant infrastructure challenge.
For example, a company running an AI chatbot must process huge numbers of requests simultaneously.
Speed matters.
Energy efficiency matters.
The amount of computing required per request matters.
This creates an opportunity for specialised processors designed specifically for inference.
Rather than relying exclusively on general-purpose GPUs, data-centre operators can potentially combine different types of chips depending on the workload.
A Startup With Serious Backing
d-Matrix is not a newcomer operating entirely on promise.
The Santa Clara-based company has attracted significant investment, including backing from Microsoft.
It raised $110 million in 2023 and later reached a valuation of about $2 billion following a $450 million funding round in 2025. The company shipped its first AI chip in late 2024.
Its latest Nvidia partnership therefore represents another step in the commercialisation of a company that has already attracted major technology investors.
The Bigger AI Infrastructure Race
The d-Matrix announcement reflects how competitive the AI hardware market has become.
Nvidia remains the dominant force in AI computing, but specialised chip startups are trying to capture specific parts of the market.
Some are targeting training.
Others are focusing on inference.
Some are building chips for robotics, autonomous vehicles or edge computing.
This fragmentation could ultimately benefit AI developers because different workloads may no longer need to rely on exactly the same hardware.
For Nvidia, partnerships such as this could also strengthen its position by making its networking and server ecosystem compatible with a broader range of processors.
For d-Matrix, the opportunity is even bigger.
If its technology can become part of mainstream AI infrastructure, the company could move from being a specialist chip startup to becoming an important component of the next generation of AI data centres.
